aboutsummaryrefslogtreecommitdiff
AgeCommit message (Collapse)Author
2019-08-30Start to use vectorizable indexing for material initializationAdrian Kummerlaender
`utility.ndindex` adds support for specifying the traversal order to `numpy.ndindex`.
2019-07-25Dampen channel inflowAdrian Kummerlaender
2019-07-18Update README.mdAdrian Kummerlaender
2019-07-18Add another GL interop exampleAdrian Kummerlaender
…just for fun
2019-07-10Update slides for talkAdrian Kummerlaender
2019-07-10Add basic talk slidesAdrian Kummerlaender
2019-07-10Add README.mdAdrian Kummerlaender
2019-07-08Update benchmark plotsAdrian Kummerlaender
2019-07-06Update benchmark plotsAdrian Kummerlaender
2019-07-06Update benchmark scriptsAdrian Kummerlaender
2019-07-06Add further non-CSE benchmark results @ P100Adrian Kummerlaender
2019-07-04Add further non-CSE benchmark results @ K2200Adrian Kummerlaender
2019-07-04Update benchmark plotsAdrian Kummerlaender
2019-07-04Update benchmark results of LDC @ Tesla P100Adrian Kummerlaender
2019-07-02Determine discrete velocities of D2Q9 and D3Q27Adrian Kummerlaender
2019-07-02Determine lattice speed of soundAdrian Kummerlaender
2019-07-02Determine weights using Gauss-Hermite quadratureAdrian Kummerlaender
2019-07-01Expand LDC benchmark scriptsAdrian Kummerlaender
2019-06-30Move OpenCL buffers into Memory classAdrian Kummerlaender
2019-06-29Implement layout and memory paddingAdrian Kummerlaender
There are at least two distinct areas where padding can be beneficial on a GPU: 1. Padding the global thread sizes to support specific thread layouts e.g. (32,1) layouts require the global lattice width to be a multiple of 32 2. Padding the memory layout at the lowest level to align memory accesses i.e. some GPUs read memory in 128 Byte chunks and as such it is beneficial if the operations are aligned accordingly For lattice and thread layout sizes that are exponents of two these two padding areas are equivalent. However when one operates on e.g. a (300,300) lattice using a (30,1) layout, padding to 128 bytes yields a performance improvement of about 10 MLUPS on a K2200. Note that I am getting quite unsatisfied with how the Lattice class and its suroundings continue to accumulate parameters. The naming distinction between Geometry, Grid, Memory and Lattice is also not very intuitive.
2019-06-28Move some common benchmark plots into helper functionsAdrian Kummerlaender
2019-06-27Add some benchmark plotsAdrian Kummerlaender
2019-06-25Adapt benchmark results format to be importableAdrian Kummerlaender
2019-06-25Fix LDC 3D x-z-plane plotAdrian Kummerlaender
2019-06-25Add raw data of Tesla P100 benchmarksAdrian Kummerlaender
2019-06-24Add basic benchmark scripts, K2200 resultsAdrian Kummerlaender
2019-06-22Add interactive 2D LDC notebook, fix material initializationAdrian Kummerlaender
2019-06-22Add platform, precision and thread layout parametersAdrian Kummerlaender
2019-06-22Extract parameters in GL interop exampleAdrian Kummerlaender
2019-06-21Gather interop moments in a more generic mannerAdrian Kummerlaender
i.e. return unshifted moments in a implicitly ordered float4 array. Cell positions are reconstructed by a vertex shaded analogously to how it is done in compustream.
2019-06-20Prototype OpenGL interoperationAdrian Kummerlaender
2019-06-20Move back assignmentAdrian Kummerlaender
2019-06-18Expand square expressionsAdrian Kummerlaender
Yields another ~5-10 MLUPS in the simple D2Q9 example. Now averaging at ~840 MLUPS for D2Q9 and ~ 400 MLUPS for D3Q19 on a K2200.
2019-06-17Extract population offsetAdrian Kummerlaender
2019-06-17Add function for exporting moments as VTK filesAdrian Kummerlaender
2019-06-16Add PyEVTK to environmentAdrian Kummerlaender
2019-06-16Replace some explicit dimension branchingAdrian Kummerlaender
2019-06-16Select thread layout depending on the descriptor's characteristicsAdrian Kummerlaender
2019-06-16Declutter gid and offset calculationAdrian Kummerlaender
2019-06-16Add D3Q27 descriptorAdrian Kummerlaender
2019-06-15Split descriptors and symbolic formulationAdrian Kummerlaender
2019-06-15Add support for generating a D3Q19 kernelAdrian Kummerlaender
Note how this basically required no changes besides generalizing cell indexing and adding the symbolic formulation of a D3Q19 BGK collision step. Increasing the neighborhood communication from 9 to 19 cells leads to a significant performance "regression": The 3D kernel yields ~ 360 MLUPS compared to the 2D version's ~ 820 MLUPS.
2019-06-15Start to record some benchmarksAdrian Kummerlaender
2019-06-15Consistently name population buffersAdrian Kummerlaender
2019-06-14Extract geometry informationAdrian Kummerlaender
2019-06-13Further the separation between descriptor and latticeAdrian Kummerlaender
2019-06-13Tidy up symbolic kernel generationAdrian Kummerlaender
2019-06-13Add JupyterLab to environmentAdrian Kummerlaender
2019-06-13Add kernel customization point for velocity boundariesAdrian Kummerlaender
2019-06-12Port LDC example to new structureAdrian Kummerlaender